Frankly I think it could be pretty good idea to make some local cluster to store only near located users data (when I say near I mean those who are not that far in channel sense, not in sense of geographical nearness). It will save some traffic (not that much, actually), but it also decrease response time and will give some interesting thing like automatically tracking local communities: it may be easy to auto-create community out of users who are on the same cluster and thus are living pretty close to each other.
There are 2 problems I see here which are pretty unknown for me: 1) how to consider who should be on that cluster and 2) how should cluster make its sync with other LJ clusters.
I think some kind of IP-filtering will solve the first problem but anyway there is a question of moving users across clusters as user will change it's IP location (I mean, first of all we need to re-allocate user's data from some cluster to this local one as we see the user uses local IP, and if it change the IP then we should maybe move his data again to put it to another cluster). But then is there any stat of which IPs the given user used to use?
The sync problem is something I know nothing about. How much data should be synced? That's the question for me!
Maybe the decision is making some kind of local (let's call it so) proxies that will store only data that are for local audience. Then this server will only save pages but all the logic stays on main LJ servers and our local one will be just slave.
I think I can make some kind of local LJ mirror here in Kaliningrad, Russia, where I'm living. Is it possible as of the current way of LJ development? If so, what's the best way of doing it?